Table of Contents
If you are struggling to decide which language model tools to use, and which are worth paying for, you are not alone!
A few people have asked me lately which LLMs I use so I thought I'd write this post. This will be an interesting post to come back to in a few months as I expect my preferences will change as each of the tools advance.
I use LLMs constantly throughout the day. I can't imaging working without them.
For writing
I generally do not use LLMs to write content. People like reading human-written content. This post you are reading now is written by me. However, I'm currently experimenting with using Gemini to rewrite product pages for eCommerce sites.
Here is what I currently use:
Gemini in AI Studio is my go-to when it comes to writing. Choose the latest model, which at the current time is Gemini 2.0 Flash, from the dropdown. I'll use prompts like these:
- rewrite this so that it is more concise
- does this make sense? Is it accurate?
- I'm writing an article on ______. What are the main points I should cover? Write an outline for me to follow.
Why Gemini over ChatGPT? I find Gemini usually sounds more conversational and less AI-like although neither have impressed me yet to the point where I rely on them to do the majority of my writing.
I also really enjoy using NotebookLM to help me with writing. I'll put a bunch of sources in the Notebook and use it to find exact quotes that go along with whatever it is that I'm writing about. This is especially helpful if I'm learning about a new topic. For example, I have a notebook with some testimonies from the DOJ vs Google trial, the API files and more.
Winner: The latest model of Gemini in AI Studio
For optimizing content
I currently find myself using Gemini in AI studio for this as well. I used to use mostly ChatGPT for analyzing content, but lately Gemini is more helpful. Here are some things you can do with it:
- Here is my content [paste it in] and here is my competitors' [paste]. Give me 5 reasons why people may prefer their content over mine along with ideas for improvement.
- I want to rank better for the phrase _____. What is the main intent of someone who types this? What microintents do they have? Tell me how each of these pieces of content perform when it comes to meeting intent [paste]? Tell me how I can improve my content to better meet search intent?
I also have thoroughly enjoyed using Deep Research which currently is only available with a personal Gemini Advanced account. I hope to be publishing information soon on the prompts that I am using to have Gemini research competitor content and help me improve the content I am working on to make it more helpful than what currently exists.
I've also started experimenting with screensharing with Gemini in AI studio as we look at competitor content and the site I'm working on. It is a wild experience to do this! I expect I'll be using this more and more.
ChatGPT does a decent job at comparing helpfulness and coming up with ideas as well.
Winner: The latest model of Gemini in AI Studio + Gemini Advanced Deep Research
For doing things with data
I really enjoy using Advanced Data Analysis in ChatGPT. If you do any work with spreadsheets, try this prompt:
- Here is a spreadsheet. What kind of visualizations or analysis can we do with it?
Here's a GPT I created that allows you to upload GSC data following a Google update and create a scatter plot showing changes in keyword rankings or clicks to pages. You're welcome to use it! Just ask it for instructions if you're not sure how.
It's not perfect. Despite me giving it instructions to make improved keywords green and declined ones red, I often need to remind the GPT of this. And it struggles to understand lost and new keywords as you can see in the chart below. Yet, I find this very helpful. We can see that this site has had nice improvements in rankings of many, but not all keywords following the November 11 core update.
I can then ask questions about the data and converse with ChatGPT about it.
Gemini doesn't do as well at this type of task so far.
Winner: ChatGPT
For image creation
I love using Imagen in Gemini Advanced, or the image creator in Grok. Grok used to use Flux for generating images and recently it changed to its own system - Aurora. I think Flux was a little better perhaps.
I find ChatGPT looks far too AI generated. I expect they will improve though.
Winner: Gemini Imagen
For brainstorming
Both ChatGPT and Gemini in AI studio are really good here. If I had to pick one model to brainstorm with it would be o1 in ChatGPT. This is the reasoning model that thinks before coming up with an answer.
Here's my favourite way to brainstorm:
- I ramble on about my problem I'm working on, either in a Google Doc or on my voice recorder on my phone.
- I give the transcript of this rambling to o1 in ChatGPT and ask it for thoughts and improvements. I'll talk back and forth with ChatGPT for a while about this. You could also use Gemini for this as well.
- I copy that conversation into NotebookLM and have it create a podcast. I tell the podcasters, "Help me brainstorm further. Come up with even better ideas. You are speaking to me, Marie." Then I've got a personalized podcast discussing the problem I'm working on!
Winner: ChatGPT's o1
For General Knowledge
This is one that I jump from model to model for. If I want to know the answer to a question, I'll usually start with Gemini in AI studio.
If it's something that would be best used with realtime Search, I use either ChatGPT Search or, what I think is a model we need to pay close attention to, Grok in X.
Grok is particularly interesting because it draws on the web plus tweets. I asked Grok which web index it uses. Gemini obviously uses Google Search, and ChatGPT draws from Bing's index, but what about Grok? Apparently X has its own index of the web?!?
Winner: Grok
For recipes
They are all so good! Lately I find myself using Gemini in AI studio the most though. I've rarely had a problem with the recipes I've created. Try this prompt.
Give me some really interesting ways to make [whatever it is you want to make], or
Give me a recipe for _______ that is as delicious as possible.
You can then converse about it and ask for ingredient substitutes, variations and more.
Winner: Gemini in AI Studio
For coding
I am not a coder, but I have created quite a few things with the help of ChatGPT. I have so far found ChatGPT to be more helpful than Gemini although I have to admit I have not tried Gemini for coding since 2.0 came out.
If you're an SEO, I would encourage you to do what I did and use ChatGPT to teach you how to connect with the GSC API. I knew no Python. I started with a prompt like this:
I want to make something that allows me to connect with the GSC API. Can you help me build it? Start at the beginning and walk me through things one step at a time. Not too much info at once. Remember I am a beginner.
Winner: ChatGPT-4o
Why not Claude?
I think Claude is amazing. When it first came out I had a conversation with Claude about the meaning of life that truly changed my perspective on the world we live in. I have also heard that it is a better option for coding. I did try it several times for coding and found that for me, ChatGPT was better.
While I think Claude is quite good, I believe that the majority of the world will end up using Gemini, ChatGPT and Grok for their LLM use in the future. Who knows...perhaps I am wrong on that!
What if I had to pick just one LLM to use?
If I could only pay for one LLM it would be a tough call. I go back and forth on this answer between Gemini and ChatGPT.
For a long time, ChatGPT was better at everything.
However, Gemini has caught up or is better for most tasks I use LLMs for today.
Also, Gemini is Google's product, and as we learned in the DOJ vs. Google trial, Google is the default search engine on most phones. Perhaps ChatGPT powering Siri will change this. I believe that at some point users will have the choice of Gemini or ChatGPT to use with Siri.
Gemini Advanced's Deep Research is mind blowing. I keep thinking of new ways to use it.
We have not talked yet about agents. As I write this, Google has just announced Project Astra (real time video chat with Gemini - you can try it out by choosing Stream Realtime in AI Studio) and Project Mariner - an agent that will take over your Chrome browser and do stuff for you like help you return an item you bought online. And in a weird coincidence, as I was writing this Google just called me to set up an appointment to talk about using Google Agentspace - AI agents, powered by Google.
The day will come (and likely sooner than we think) where every business has an AI agent. I think we haven't started to grasp what that will look like. My personal agent which knows my likes and how I work, will interact with your business agent to book appointments, help me solve problems, sell me products and more. I imagine we are not far from the day where I can be in my garden and use Project Astra in my glasses while gardening and say, "Gemini - why is my plant a strange colour." And Gemini will say, "Oh, that looks like you have a fungus. Want me to order a product for you?" Gemini will know which product is popular and effective for my plants and area, and also is the best price, and order it and ship it to me (perhaps by drone one day?) without me ever having to visit a website.
I tell you this because there are two reasons for us to use LLMs right now. If your main reason is to use AI to help you get things done more efficiently, then I'd recommend trying both ChatGPT and Gemini with paid accounts for a few months and seeing which you click best with. Or, if you're using it for coding, I'd try out Claude as well.
However, if your goal is the same as mine - to become really good at understanding how to communicate with LLMs as they change the world, I would recommend you use Gemini. It's worth paying for Gemini Advanced to use Deep Research and to use it within your Google Docs, email and more. While OpenAI has a chance to put a dent in Google's market share, I believe Google will come out ahead. They are on every phone. They will bring capabilities to allow us to turn our knowledge in Google docs, slides and more into Agents. And I expect that this will become a new revenue stream for them as well. Instead of banner and search ads, we'll likely have ads that encourage people to connect with a business' AI agent.
With all of this being said, I would not discount X's Grok. Elon Musk disrupts every industry he enters. Grok has the biggest cluster of computers for AI training. It has access to real time information, not as published by the media, but as it is shared by real people, and as video data becomes training data, it will have information from every Tesla and eventually, every Optimus Robot and who knows what kind of information X is learning from Starlink! When one car or robot learns something, each one in the fleet will. X will grow quickly.
I think the race to become the AI everyone uses is not between OpenAI and Google, but rather between Google and XAI.
I would encourage you to try using an LLM every day. You will find it is not perfect. Criticizing an LLM for its mistakes is like talking about a toddler saying, "Well sure, they can do some cute stuff, but when I asked them this question they failed." LLM's will continue to improve. We are still in early days.
We have not even started to talk about what happens as we integrate AI into our brains. I know that sounds like crazy talk, but I am certain we have already started to merge with machines. Can you imagine living life without your phone?
I hope you've found this helpful. If you want to stay up to date on the latest interesting stories on LLMs, AI and Search, here is my newsletter:
Marie.